2 * Assembly testing and benchmarking tool
3 * Copyright (c) 2015 Henrik Gramner
4 * Copyright (c) 2008 Loren Merritt
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License along
19 * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
20 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
23 #ifndef TESTS_CHECKASM_CHECKASM_H
24 #define TESTS_CHECKASM_CHECKASM_H
28 #include "libavutil/avstring.h"
29 #include "libavutil/cpu.h"
30 #include "libavutil/lfg.h"
31 #include "libavutil/timer.h"
33 void checkasm_check_alacdsp(void);
34 void checkasm_check_bswapdsp(void);
35 void checkasm_check_flacdsp(void);
36 void checkasm_check_fmtconvert(void);
37 void checkasm_check_h264pred(void);
38 void checkasm_check_h264qpel(void);
39 void checkasm_check_jpeg2000dsp(void);
40 void checkasm_check_pixblockdsp(void);
41 void checkasm_check_synth_filter(void);
42 void checkasm_check_v210enc(void);
43 void checkasm_check_vp9dsp(void);
44 void checkasm_check_videodsp(void);
46 void *checkasm_check_func(void *func, const char *name, ...) av_printf_format(2, 3);
47 int checkasm_bench_func(void);
48 void checkasm_fail_func(const char *msg, ...) av_printf_format(1, 2);
49 void checkasm_update_bench(int iterations, uint64_t cycles);
50 void checkasm_report(const char *name, ...) av_printf_format(1, 2);
52 /* float compare utilities */
53 int float_near_ulp(float a, float b, unsigned max_ulp);
54 int float_near_abs_eps(float a, float b, float eps);
55 int float_near_abs_eps_ulp(float a, float b, float eps, unsigned max_ulp);
56 int float_near_ulp_array(const float *a, const float *b, unsigned max_ulp,
58 int float_near_abs_eps_array(const float *a, const float *b, float eps,
60 int float_near_abs_eps_array_ulp(const float *a, const float *b, float eps,
61 unsigned max_ulp, unsigned len);
63 extern AVLFG checkasm_lfg;
64 #define rnd() av_lfg_get(&checkasm_lfg)
66 static av_unused void *func_ref, *func_new;
68 #define BENCH_RUNS 1000 /* Trade-off between accuracy and speed */
70 /* Decide whether or not the specified function needs to be tested */
71 #define check_func(func, ...) (func_ref = checkasm_check_func((func_new = func), __VA_ARGS__))
73 /* Declare the function prototype. The first argument is the return value, the remaining
74 * arguments are the function parameters. Naming parameters is optional. */
75 #define declare_func(ret, ...) declare_new(ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
76 #define declare_func_emms(cpu_flags, ret, ...) declare_new_emms(cpu_flags, ret, __VA_ARGS__) typedef ret func_type(__VA_ARGS__)
78 /* Indicate that the current test has failed */
79 #define fail() checkasm_fail_func("%s:%d", av_basename(__FILE__), __LINE__)
81 /* Print the test outcome */
82 #define report checkasm_report
84 /* Call the reference function */
85 #define call_ref(...) ((func_type *)func_ref)(__VA_ARGS__)
87 #if ARCH_X86 && HAVE_YASM
88 /* Verifies that clobbered callee-saved registers are properly saved and restored
89 * and that either no MMX registers are touched or emms is issued */
90 void checkasm_checked_call(void *func, ...);
91 /* Verifies that clobbered callee-saved registers are properly saved and restored
92 * and issues emms for asm functions which are not required to do so */
93 void checkasm_checked_call_emms(void *func, ...);
96 /* Evil hack: detect incorrect assumptions that 32-bit ints are zero-extended to 64-bit.
97 * This is done by clobbering the stack with junk around the stack pointer and calling the
98 * assembly function through checked_call() with added dummy arguments which forces all
99 * real arguments to be passed on the stack and not in registers. For 32-bit arguments the
100 * upper half of the 64-bit register locations on the stack will now contain junk which will
101 * cause misbehaving functions to either produce incorrect output or segfault. Note that
102 * even though this works extremely well in practice, it's technically not guaranteed
103 * and false negatives is theoretically possible, but there can never be any false positives.
105 void checkasm_stack_clobber(uint64_t clobber, ...);
106 #define declare_new(ret, ...) ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__)\
107 = (void *)checkasm_checked_call;
108 #define declare_new_emms(cpu_flags, ret, ...) \
109 ret (*checked_call)(void *, int, int, int, int, int, __VA_ARGS__) = \
110 ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \
111 (void *)checkasm_checked_call;
112 #define CLOB (UINT64_C(0xdeadbeefdeadbeef))
113 #define call_new(...) (checkasm_stack_clobber(CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,\
114 CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB,CLOB),\
115 checked_call(func_new, 0, 0, 0, 0, 0, __VA_ARGS__))
117 #define declare_new(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call;
118 #define declare_new_emms(cpu_flags, ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = \
119 ((cpu_flags) & av_get_cpu_flags()) ? (void *)checkasm_checked_call_emms : \
120 (void *)checkasm_checked_call;
121 #define call_new(...) checked_call(func_new, __VA_ARGS__)
123 #elif ARCH_ARM && HAVE_ARMV5TE_EXTERNAL
124 /* Use a dummy argument, to offset the real parameters by 2, not only 1.
125 * This makes sure that potential 8-byte-alignment of parameters is kept the same
126 * even when the extra parameters have been removed. */
127 void checkasm_checked_call_vfp(void *func, int dummy, ...);
128 void checkasm_checked_call_novfp(void *func, int dummy, ...);
129 extern void (*checkasm_checked_call)(void *func, int dummy, ...);
130 #define declare_new(ret, ...) ret (*checked_call)(void *, int dummy, __VA_ARGS__) = (void *)checkasm_checked_call;
131 #define call_new(...) checked_call(func_new, 0, __VA_ARGS__)
132 #elif ARCH_AARCH64 && !defined(__APPLE__)
133 void checkasm_checked_call(void *func, ...);
134 #define declare_new(ret, ...) ret (*checked_call)(void *, __VA_ARGS__) = (void *)checkasm_checked_call;
135 #define call_new(...) checked_call(func_new, __VA_ARGS__)
137 #define declare_new(ret, ...)
138 #define declare_new_emms(cpu_flags, ret, ...)
139 /* Call the function */
140 #define call_new(...) ((func_type *)func_new)(__VA_ARGS__)
143 #ifndef declare_new_emms
144 #define declare_new_emms(cpu_flags, ret, ...) declare_new(ret, __VA_ARGS__)
147 /* Benchmark the function */
149 #define bench_new(...)\
151 if (checkasm_bench_func()) {\
152 func_type *tfunc = func_new;\
155 for (ti = 0; ti < BENCH_RUNS; ti++) {\
156 uint64_t t = AV_READ_TIME();\
161 t = AV_READ_TIME() - t;\
162 if (t*tcount <= tsum*4 && ti > 0) {\
167 checkasm_update_bench(tcount, tsum);\
171 #define bench_new(...) while(0)